Institutional members access full text with Ovid®

The Multitime Case-control Design for Time-varying Exposures

Suissa, Samya,b; Dell'Aniello, Sophieb; Martinez, Carlosb

doi: 10.1097/EDE.0b013e3181f2f8e8
Methods: Original Article

Background: The conventional approach to improve precision of the odds ratio in a case-control study is to increase the number of controls per case. With time-varying exposures, an alternative is to increase the number of observations per control.

Method: We present the multitime case-control design, which uses multiple control person-moments of exposure within each control subject. The point and variance estimators of the odds ratio are corrected for within-subject correlation. We illustrate this approach using case-control data from studies of the effects of respiratory medications.

Results: Simulations show that, with uncorrelated exposures, it is possible to reduce the variance of the odds ratio by around 30% by increasing the number of control person-moments per subject. With correlated exposures, an accurate variance can be obtained by correcting for within-subject correlation. The corrected variance increases with increasing correlation, depending on the number of control person-moments. The first illustration shows that the rate ratio (RR) of cardiac death associated with β-agonist use, not estimable with 1 control per case (30 cases) and 1 control person-moment, was 4.2 (95% confidence interval = 0.4–49) with 12 control person-moments. The second example finds a rate ratio of acute myocardial infarction associated with antibiotics of 2.00 (1.16–3.44) with 1 control per case, which improves in precision with 10 control subjects per case (RR = 2.13 [1.48–3.05]) but also with 1 control per case and 10 control person-moments per control subject (1.99 [1.36–2.90]).

Conclusion: When dealing with time-varying exposures, the multitime case-control design can increase the efficiency of conventional case-control studies without additional control subjects.

From the aDepartment of Epidemiology and Biostatistics, McGill University, Montreal, Canada; and bMcGill Pharmacoepidemiology Research Unit, Centre for Clinical Epidemiology, Lady Davis Research Institute, Jewish General Hospital, Montreal, Canada.

Submitted 15 May 2009; accepted 7 June 2010.

Supported by the Canadian Institutes of Health Research (CIHR).

Correspondence: Samy Suissa, Centre for Clinical Epidemiology, Lady Davis Research Institute, Jewish General Hospital, 3755 Cote Ste-Catherine, Montreal, Québec, Canada H3T 1E2. E-mail:

© 2010 Lippincott Williams & Wilkins, Inc.